# EU AI Act Article 50: is your AI persona a deepfake? EU AI Act Article 50 has applied since 2 August 2026. Its deepfake rule says you have to label a photorealistic AI persona as AI. What counts, and the fines. - Canonical: https://aiofmcity.com/blog/eu-ai-act-article-50-persona-operators - Topic: EU compliance (https://aiofmcity.com/topics/eu-compliance) - Author: Gus - Published: 2026-09-07 - Updated: 2026-10-05 --- ## Key points - Since 2 August 2026, if you publish photorealistic AI-generated people in the EU, you have to say that the content is artificially generated. - The Commission's guidelines of 20 July 2026 say you have to label the content even if no real person is shown. - A line at the bottom of your bio is not enough. People have to see the label the first time they see the content. - Article 50 has no minimum revenue. It applies to everyone who publishes, however small the business is. - In Germany § 184b StGB covers sexual depictions of children, including realistic AI-generated ones. No real child has to be involved. --- > **This is not legal advice, and I am not a lawyer.** I read the legal texts named in each source box below and wrote down what they say. Use this article to know which questions to ask a qualified IT and media lawyer. The fines are high, so one paid hour with a real lawyer is the cheapest cost in this business. If you publish photorealistic AI-generated images, video or audio of a person, EU law says you have to label the content as artificially generated. This has applied since 2 August 2026. It also applies to a persona who does not exist. The rest of this article explains what "photorealistic", "label" and "you" mean in practice, because those are the points people get wrong. ## What Article 50(4) covers Article 50 is in the transparency chapter of Regulation (EU) 2024/1689, the AI Act. Its rules apply to *deployers*. A deployer is the person who publishes the output, so the rules do not only apply to the company that built the model. If you run the account, you are the deployer. > Deployers of an AI system that generates or manipulates image, audio or video content > constituting a deepfake must disclose that the content has been artificially generated or > manipulated. Separately, deployers of AI systems that generate or manipulate text published > to inform the public on matters of public interest must disclose that the text is > artificially generated, unless the content underwent human review and a natural or legal > person holds editorial responsibility for it. > > — Source: EU AI Act Art. 50(4), read 2026-09-08 — https://eur-lex.europa.eu/eli/reg/2024/1689/oj There are two rules, and they apply in different cases. The rule for images, audio and video applies to a persona account. The rule for text applies to anyone who runs an AI-assisted blog, including this one. The key word is "deepfake". In the law it does not mean what it means in everyday talk. > "Deepfake" means AI-generated or manipulated image, audio or video content that resembles > existing persons, objects, places, entities or events and would falsely appear to a person > to be authentic or truthful. > > — Source: EU AI Act Art. 3(60), read 2026-09-08 — https://eur-lex.europa.eu/eli/reg/2024/1689/oj The definition says nothing about a named victim, nothing about intent and nothing about harm. It asks two questions. Does the content look like real persons, objects, places or events? Would a viewer think it is a photograph? For a studio-lit portrait of an invented 24-year-old in a real-looking apartment, the answer to both is yes. ## Why Article 50 also applies to an invented persona This point costs people the most time. Many think: *she is not a real person, so there is nobody to deceive.* That feels right, but it is wrong. The Commission clearly says the rule still applies. > The guidelines confirm that the labelling obligation applies even in the absence of > deceptive intent, and even where the content does not depict an identifiable real person. > Content that is clearly fantastical, stylised or physically impossible falls outside the > duty; photorealistic depictions of humans do not. > > — Source: European Commission guidelines on Article 50 transparency obligations, read 2026-09-08 — https://digital-strategy.ec.europa.eu/en/policies/guidelines-ai-transparency-obligations An AI persona earns money because her images look like photographs of a person. That same quality puts the images inside Article 50(4). You cannot make photorealistic humans in this business and stay outside the rule. Plan for the rule and do not look for a way around it. There is an exception, but it is narrow. Anime, an obvious 3D-render look, stylised illustration, anything a viewer would not mistake for a camera photo: all of that is outside the rule. If your whole account looks like that, this article is background reading for you. If your account mixes both, only the photoreal posts need the label. In practice it is easier to label everything than to decide for every single post. ## The rule for AI-generated text The text part of Article 50(4) applies in fewer cases. It covers text "published with the purpose of informing the public on matters of public interest". There is an exception for content that a human reviewed and for which someone has editorial responsibility. The Commission reads that exception strictly. > A superficial pass over model output does not satisfy the human-review exception. > The guidelines require substantive editorial control with a clearly identifiable person or > entity accountable for the published text. > > — Source: European Commission guidelines on Article 50 transparency obligations, read 2026-09-08 — https://digital-strategy.ec.europa.eu/en/policies/guidelines-ai-transparency-obligations For a persona account this rarely matters. Captions and DMs are not published to inform the public on matters of public interest. If you also run a blog about platform rules and regulation, the text rule probably applies. The cheapest answer is to skip the argument about whether your content counts. Keep a named author, a review step before you publish, and a dated record of who approved what. A pull-request history gives you that record for free. ## What you have to change when you post Three things change. The one that goes wrong most often comes first. - **Label it where people first see it.** The label has to be clear and visible when someone first sees the content. It is not enough if they can find it later. In practice, put a visible marker on or next to the post itself, and switch on the platform's own AI label where there is one. A line at the end of your bio does not count. The [AI disclaimer generator](https://aiofmcity.com/ai-disclaimer-generator) gives you the platform setting and a line to copy for each platform. - **Keep the machine-readable markers.** These are the C2PA manifests and watermarks that the generator embeds. The AI Act does not ban you from removing them. Keep them anyway: platforms read them to label your posts, and your generator's terms may forbid removing them. Normal image and video tools throw that data away when they re-encode a file. There is a separate article about it: [why re-encoding pipelines silently strip your provenance](https://aiofmcity.com/blog/dont-strip-your-watermarks-provenance). - **Keep a record.** Write down which model, which date, which account and which label. The regulation does not set a log format. But "we always labelled" is only a claim, and a dated export is proof. ### What did not work: Putting the AI disclosure as the last line of the Instagram bio It looked like it followed the rule, but it did not. The label sat three taps away from the content it described. A viewer who saw a Reel in their recommendations never saw the label. It also did nothing for the same images cross-posted to Reddit and X, because the bio is not shown with the post. Replacing it with a caption line on every post plus the platform's own AI-generated profile label took about forty minutes of back-editing. After that there was nothing left to argue about. Compared with the previous four weeks, there was no measurable loss of reach. ## The AI Act, Instagram and Fanvue ask for the same label Three separate sets of rules ask you to do the same thing. > Instagram replaced its "AI creator" label with "AI-generated profile" on 31 August 2026. > Accounts presenting an AI-generated identity without the label receive reduced distribution > on recommendation surfaces to non-followers. Accounts that use AI only as an editing or > captioning tool are not covered. > > — Source: Instagram Creators — AI-generated profile label, read 2026-09-08 — https://creators.instagram.com/blog/ai-generated-profile-label > Fanvue requires disclosure of AI-generated media — in the profile bio, the content caption, or as a watermark — and its AI-content policy prohibits AI-generated media depicting any real person other than the account owner, including anyone who resembles or is intended to resemble someone under 18. A prompt or disclaimer cannot be used to establish age; visual appearance is what is judged. > > — Source: Fanvue Help — is AI content allowed on Fanvue, read 2026-09-08 — https://help.fanvue.com/en/articles/9538738-is-ai-content-allowed-on-fanvue For verification you use your own identity document, not the persona's. One way of labelling meets the AI Act, Instagram's reach rule and Fanvue's terms at the same time. Make the label a normal part of how you post. Then the compliance work comes down to one caption template and one decision about metadata. ## What Article 50 does not cover Four things are outside the rule. Know them exactly. If you are vague here, you either stop posting out of fear or feel safe when you are not. - **Content generated *and* published before 2 August 2026.** You do not have to label it afterwards. Both conditions must be true: an old image first posted in September 2026 counts as newly published. - **Clearly fantastical or impossible content.** Stylised, obviously synthetic, not photoreal. - **AI as a tool and not as the identity.** Colour grading a real photograph, generating captions, upscaling. Instagram's rule draws the same line: what counts is whether the *depicted identity* is AI-generated. - **Purely private use.** Article 50 applies when you publish or share content. It does not apply to images that never leave your drive. ## Article 50 fines and who they apply to > Non-compliance with the transparency obligations in Article 50 is subject to administrative > fines of up to EUR 15 000 000 or, for undertakings, up to 3% of total worldwide annual > turnover for the preceding financial year, whichever is higher. > > — Source: EU AI Act Art. 99(4), read 2026-09-08 — https://eur-lex.europa.eu/eli/reg/2024/1689/oj For small businesses, including a one-person business, the lower of the two amounts is the limit: > In the case of SMEs, including start-ups, each fine referred to in this Article shall be up to the percentages or amount referred to in paragraphs 3, 4 and 5, whichever thereof is lower. > > — Source: EU AI Act Art. 99(6), read 2026-10-04 — https://eur-lex.europa.eu/eli/reg/2024/1689/oj If you run an account alone, you are not the main target of the authorities, and you will not see the maximum fine. But Article 50 has no minimum revenue. It applies as soon as you publish, however small your business is. National market surveillance authorities enforce it, and a competitor can file a complaint for free. ## § 184b StGB: AI images that look like a minor One limit in this field has nothing to do with administrative fines. Take it very seriously. > In Germany, § 184b StGB covers sexual depictions of children, including computer-generated > and AI-generated depictions where they appear realistic. No real child > needs to be involved for the provision to be engaged. Clearly stylised depictions are > treated differently; photorealistic AI images are the opposite of stylised. > > — Source: § 184b StGB (Germany), read 2026-10-04 — https://www.gesetze-im-internet.de/stgb/__184b.html In this niche, photorealistic images are what you sell. Here that makes the risk bigger. A prompt or a disclaimer does not change how old a person looks in an image. The image is judged by the impression it gives. Your intent does not count. That means: no persona look aimed at youthfulness, no "barely legal" framing, and no face LoRAs deliberately trained to look young. Do a documented check of how old the person looks before every publication. If in doubt, do not publish. There is no grey area here. ## A checklist for this week If you prefer to answer six questions, use the [self-check](https://aiofmcity.com/eu-ai-act-check). It lists the rules that apply to your setup. Where the law is not settled, it says so and does not guess. 1. Add a visible AI label to the caption template you use, on every platform. 2. Switch on the platform's own AI label where there is one: Instagram's AI-generated profile setting, Fanvue's disclosure fields. 3. Open one exported file from your pipeline and check whether the C2PA manifest is still there. 4. Write down the date each persona account started. Then you have it in writing for the 2 August 2026 cut-off and do not have to remember it. 5. Book the hour with a lawyer. Bring this checklist with you. Labelling rules and payout rules belong together. Read a platform's labelling terms and its payout terms on the same day. [How to read a platform payout page](https://aiofmcity.com/blog/reading-platform-payout-terms) shows you how. If a platform decides you got it wrong, read [risk, bans and recovery](https://aiofmcity.com/guides/risk-bans-and-recovery). You will find the other parts of this topic on the [EU compliance hub](https://aiofmcity.com/topics/eu-compliance): the Article 3(60) question, the German business setup, and the comparison of Instagram and Fanvue. ## Frequently asked ### Does Article 50 apply if my persona is not based on a real person? Yes. The Commission guidelines of 20 July 2026 say you have to label the content even if no real person is shown. This also applies if you do not mean to deceive anyone. A photorealistic invented persona is covered. Content that is clearly fantasy or physically impossible is not. ### Do I have to go back and label everything I posted before August 2026? No. You do not have to go back and label content that was both generated and published before 2 August 2026. If you regenerate, re-edit or repost it after that date, treat it as new content. ### Is a line at the bottom of my bio enough? No. The label has to be clear and visible the first time someone sees the content. A label that the viewer only finds after tapping through to your profile and scrolling does not count. A marker that is only in the file metadata does not count either. ### What is the maximum fine under Article 50? Up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher, under Article 99(4) of the AI Act. For small and medium-sized businesses, including one-person businesses, the lower of the two is the limit (Article 99(6)). The rule applies however little you earn. A single person running an account is unlikely to see the maximum fine. ### Does labelling hurt my reach? On Instagram it has been the other way round since 31 August 2026. Profiles that show an AI-generated identity without the AI-generated profile label are shown less to non-followers. You need the label to get normal reach.